README.md

---
license: apache-2.0
---
# Earth-2 Checkpoints: DLESyM-V1-ERA5

## Description:

DLESyM-V1-ERA5 is an ensemble forecast model for global earth system modeling.
This model includes an atmosphere and ocean component, using atmospheric
variables as well as the sea-surface temperature on a HEALPix nside=64
(approximately 1 degree) resolution grid. This model package includes several
individual trained checkpoints for the atmosphere and ocean components, which
can be used to improve model variability in ensembles. The model architecture
is a U-Net with padding operations modified to support using the HEALPix grid.

For training recipies see [PhysicsNeMo](https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/dlwp_healpix), for inference [Earth2Studio](https://nvidia.github.io/earth2studio/examples/14_dlesym_example.html#sphx-glr-examples-14-dlesym-example-py).

This model is ready for commercial/non-commercial use.

### License/Terms of Use: 
**Governing Terms**: Use of this model is governed by the [NVIDIA Community Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-community-models-license/).

### Deployment Geography:
Global

### Use Case:
Industry, academic, and government research teams interested in subseasonal-to-seasonal weather forecasting, and climate modeling.

### Release Date:
NGC 05/12/2025

## Reference:

 - [Advancing Parsimonious Deep Learning Weather Prediction using the HEALPix Mesh](https://arxiv.org/abs/2311.06253) <br> 
 - [A Deep Learning Earth System Model for Efficient Simulation of the Observed Climate](https://arxiv.org/abs/2409.16247) <br>

## Model Architecture: 

**Architecture Type:** DLESyM uses two UNet architectures adapted to the HEALPix grid,
one for each of the atmosphere and ocean components.  <br>
**Network Architecture:** UNet <br>

## Input:

**Input Type:**

- Tensor (9 surface and pressure-level variables)

**Input Format:** PyTorch Tensor  <br>
**Input Parameters:**

- Six Dimensional (6D) (batch, lead time, variable, face, height, width) <br>

**Other Properties Related to Input:** 

- Input latitude/longitude grid: 0.25 degree 721 x 1440, regridded to HEALPix nside=64
grid in "XY" format with a north origin and clockwise order
  - See `HEALPIX_PAD_XY` in the [earth2grid package](https://github.com/NVlabs/earth2grid)
    for specific details
- Input state weather variables: `z500`, `tau300-700`, `z1000`, `t2m`, `tcwv`, `t850`,
`z250`, `ws10m`, `sst`
  - `tau300-700` (geopotential thickness) is defined as the difference between `z300`
  and `z700` geopotential levels.
  - `ws10m` (wind speed at 10m above surface) is defined as the square root of the sum
  of the squared zonal and meridional wind components, i.e. `sqrt(u10m **2 + v10m **2)`.

For variable name information, review the `HRRR` Lexicon at [Earth2Studio](https://github.com/NVIDIA/earth2studio).
Review the `config.yaml` provided in the model package for information on the input lead
times required by the model.

## Output:

**Output Type:** Tensor (9 surface and pressure-level variables) <br>
**Output Format:** Pytorch Tensor <br>
**Output Parameters:** Six Dimensional (6D) (batch, lead time, variable, face, height,
width)  <br>
**Other Properties Related to Output:**

- Output latitude/longitude grid: 0.25 degree 721 x 1440, regridded to HEALPix nside=64
grid in "XY" format with a north origin and clockwise order.
  - See `HEALPIX_PAD_XY` in the [earth2grid package](https://github.com/NVlabs/earth2grid)
    for specific details
- Output state weather variables: `z500`, `tau300-700`, `z1000`, `t2m`, `tcwv`, `t850`,
`z250`, `ws10m`, `sst`   <br>
  - `tau300-700` (geopotential thickness) is defined as the difference between `z300`
  and `z700` geopotential levels.
  - `ws10m` (wind speed at 10m above surface) is defined as the square root of the sum
  of the squared zonal and meridional wind components, i.e. `sqrt(u10m **2 + v10m **2)`.

Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems.
By leveraging NVIDIA’s hardware (e.g. GPU cores) and software frameworks (e.g., CUDA
libraries), the model achieves faster training and inference times compared to
CPU-only solutions.

## Software Integration

**Runtime Engine:** Pytorch <br>
**Supported Hardware Microarchitecture Compatibility:** <br>
* NVIDIA Ampere <br>
* NVIDIA Hopper <br>
* NVIDIA Turing <br>


**Supported Operating System:**
* Linux <br>


## Model Version: 

**Model Version:** v1 <br>

# Training, Testing, and Evaluation Datasets: 

**Total size (in number of data points):** 110,960 <br>
**Total number of datasets:** 1<br>
**Dataset partition:** training 90%, testing 5%, validation 5% <br>

## Training Dataset:

**Link:** [ERA5](https://cds.climate.copernicus.eu/) <br>

**Data Collection Method by dataset** <br>
* Automatic/Sensors <br>

**Labeling Method by dataset** <br>
* Automatic/Sensors <br>

**Properties:**
ERA5 data for the period 1980-2015. ERA5 provides hourly estimates of various
atmospheric, land, and oceanic climate variables. The data covers the Earth on a 30km
grid and resolves the atmosphere at 137 levels. We re-grid to a healpix-64 grid that
corresponds to an approximate 1 degree lat/lon grid. <br>


## Testing Dataset:

**Link:** [ERA5](https://cds.climate.copernicus.eu/) <br>

**Data Collection Method by dataset** <br>

* Automatic/Sensors <br>

**Labeling Method by dataset** <br>
* Automatic/Sensors <br>

**Properties:**
ERA5 data for the period 2016-2017. ERA5 provides hourly estimates of various
atmospheric, land, and oceanic climate variables. The data covers the Earth on a 30km
grid and resolves the atmosphere at 137 levels. We re-grid to a healpix-64 grid that
corresponds to an approximate 1 degree lat/lon grid. <br>

## Evaluation Dataset:

**Link:** [ERA5](https://cds.climate.copernicus.eu/) <br>

**Data Collection Method by dataset** <br>
* Automatic/Sensors <br>

**Labeling Method by dataset** <br>
* Automatic/Sensors <br>

**Properties:**
ERA5 data for the period 2018-2019. ERA5 provides hourly estimates of various
atmospheric, land, and oceanic climate variables. The data covers the Earth on a 30km
grid and resolves the atmosphere at 137 levels. We re-grid to a healpix-64 grid that
corresponds to an approximate 1 degree lat/lon grid. <br>

## Inference:

**Acceleration Engine:** PhysicsNeMo, PyTorch <br>
**Test Hardware:**
* A100 <br>
* H100 <br>
* L40S <br>


## Ethical Considerations:

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications.  When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.  

Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).